[Forum SIS] DEI - Università di Catania: AVVISO DI SEMINARIO

[antonio punzo]

********************************************************************************
AVVISO DI SEMINARIO
********************************************************************************

Il Dipartimento di Economia e Impresa dell’Università di Catania ha
organizzato il seminario:



*Dealing with non-normal errors in linear regression models: a solution
based on mixtures*

*Giuliano Galimberti*

*Dipartimento di Scienze Statistiche "Paolo Fortunati"*

*University of Bologna*







Mercoledì 18 Maggio dalle ore 16:00

In Aula C, Palazzo Fortuna



Università di Catania

Palazzo delle Scienze, Corso Italia, 55, Catania





*Abstract*:



Linear regression models are well-known tools that are widely used for
studying the dependence of a (univariate or multivariate) response from a
set of regressors. One of the basic assumptions behind this class of models
is that the error terms have a (univariate or multivariate) normal
distribution. However, in many disciplines scientific research based on
empirical studies or theoretical reasoning provide support for the presence
of skewness, heavy tails and/or multimodality in the distribution of the
error terms. In order to tackle this issue, several extensions of linear
regression models have been proposed in the literature. This talk focuses
on the use of Gaussian mixtures for modelling the distribution of the error
terms. Linear regression models are studied under this assumption. First of
all, identifiability conditions are provided.  Secondly, an
Expectation-Maximization (EM) algorithm for model parameter estimation is
described, with closed form expressions for both E- and M-steps.
Furthermore, other inferential issues are addressed, such as standard error
estimates and model selection. Finally, the usefulness of the proposed
method and a numerical evaluation of its properties are illustrated through
the analysis of simulated and real datasets. Particular emphasis is given
to the ability of recovering omitted/unobserved categorical regressors,
whose categories can be related to clusters of units.





I colleghi interessati sono cordialmente invitati a partecipare.





-- 
Antonio Punzo, Ph.D.
Associate Professor of Statistics
University of Catania
Department of Economics and Business
Corso Italia 55, 95129 Catania - Italy

Office: +39 095 7537640
e-mail: antonio.punzo a unict.it
URL: http://www.economia.unict.it/punzo
Skype: antoniopunzo81
Web: google scholar
<http://scholar.google.com/citations?user=WmIbxWkAAAAJ&hl=en>
-------------- parte successiva --------------
Un allegato HTML è stato rimosso...
URL: <http://www.stat.unipg.it/pipermail/sis/attachments/20160509/8ee65795/attachment-0001.html>